MetaX Weekly AI Trend Report Week 28 2025 -- xAI Grok 4 Overtakes OpenAI and Google to Retake AI Lead, Open-Source Models Also Breaking Commercial Service Performance, AI Browser War Intensifying with Agent Marketplace Expansion, Fierce Billions of Dollars Investment Competition: Grok 4 emergence and AI frontier competition is entering a new phase. Early access Grok 4 benchmark test results reveal this model surpasses OpenAI o3 and Google Gemini 2.5 Pro, set to be the leading AI frontier. According to leaked benchmarks, Grok 4 recorded 45% score on Humanity Last Exam demonstrating cutting-edge performance. Open-source models are also showing notable developments. Hugging Face SmolLM3 was launched as a fully open 3B parameter language model supporting 6 languages with strong reasoning functions and long context processing. The OLMo 2 model family also launched as a fully open language model, with OLMo 2 32B particularly showing competitive performance with GPT-3.5 on popular benchmark suites. The AI browser war: multiple companies are now building AI-native browsers (Perplexity Comet, The Browser Company Dia, Arc browser) that embed AI assistance directly into the browsing experience rather than as a separate tool; the browser is the interface where most knowledge work happens and controlling it creates significant AI distribution advantage.